AI Model Efficiency Toolkit (AIMET)
A software toolkit from Qualcomm for quantizing and compressing trained neural network models to reduce their compute and memory footprint for edge deployment.
AIMET, the AI Model Efficiency Toolkit, is a software toolkit for quantizing and compressing trained machine-learning models so they run faster and fit into less memory on deployment hardware such as mobile phones and laptops. It supports models from the PyTorch and ONNX frameworks and applies post-training and fine-tuning techniques, including data-free quantization, to limit the accuracy loss that quantization can introduce.
Post-training quantization techniques it provides include basic calibration to compute quantization parameters, along with more advanced methods such as sequence-based MSE optimization and adaptive rounding, and it supports advanced structured and unstructured model-compression techniques alongside quantization. Models quantized with AIMET are published as part of a related collection of optimized, quantized models from Qualcomm.
AIMET is written in Python; its packages are published on PyPI, and it can also be built from source using Docker.